Post

Wait, so the recommended limit for CBOR record size is 1MiB but the hard limit for record blocks in sync 1.1 commits is 1MB. So you can write the record but the relays will choke on it atproto.com/specs/sync#c... atproto.com/guides/data-...

@bad-example.com ever run into this?

ohhhh yeah yes. or at least, it's on my list of spec things to raise on the ietf mailing list. 1MiB hard limit is for streaming #commit payloads. ietf draft text says > Repository updates exceeding these limits MUST be communicated through #sync message instead so bigger records aren't disallowed

they just need an explicit resync by sync1.1 today. but i actually think they should just be disallowed and we should enforce 1MiB everywhere (cbor, not sure if anyone has calculated the worst-case expansion size for json). arbitrarily-big: sketchy. app-specific limits: complicate sync1.1 handling

that language in the text isn't clear. big records should be disallowed. the '#sync' trick is for large batches of record ops

we would want this situation to heal/recover better too, not leave PDS sync in a bad state

Yeah, it's not very discoverable. If the relay had backoff it could've taken forever for me to realize. My alerting saying the server is pinned at 100% at least let me know about the issue!